Latest News.

AI, Language and Healthcare

The links between them and the effect of the pandemic

Artificial intelligence is everywhere 

The influence of artificial intelligence (AI) on our world is increasing at a rapid rate. Every day we use AI in our daily routines, often unknowingly. We opt for facial recognition to open our phones, we bank online and chat to a bot about our spending habits, we check our social media feeds personalised by AI and we talk to our AI-driven home assistants to ask for the latest weather report. Many of these tasks involve the processing of human language and AI’s ability to understand the way we communicate is improving steadily. 

AI and language 

Language is at the heart of AI technology. Natural language processing (NLP) is a field of artificial intelligence that focuses on giving computers the ability to comprehend, interpret and reproduce human spoken and written communication. This field has received significant attention and investment over recent years as the development of humanmachine interaction opens up a myriad of possible applications.  

NLP advances have been made possible by the availability big data, more powerful computers and advanced algorithms which in combination enable NLP to interpret and produce human language. Easy right? Well, no, pretty tricky actually and with the progress made it’s easy to overlook the difficulties. There are nuanced and illogical elements to language that make it difficult for AI to grasp; grammar, irony, humour, homophones, metaphors, to name just a few, all make NLP’s task a challenge. If your Alexa fails to understand your questions or you wonder how your GPS navigator can mispronounce a road name so badly, you will have experienced its imperfection. As humans we forget the extreme complexity of our language – we use it without thinking. 

But NLP is gradually overcoming these obstacles and our unwittingly adoption of it into our daily lives is just the beginning. The coming years will see huge advances in life enhancing AI and NLP will be a driving force. 

AIhealthcare and coronavirus 

It’s only natural that as AI technologies start to help humans live easier and more tech-based lives, the part they play in the realm of healthcare is becoming significant. An already fast-paced, innovative and high investment industry is being propelled into the future by the coronavirus crisis, and as we have previously mentioned in our blog posts, the pandemic is accelerating digitalisation in medicine. As patient consultations are pushed to a virtual setting and healthcare operators grapple with reaching overwhelming numbers of people in a safer and more efficient way, the adoption of for example, telemedicine, healthcare apps and robotics, has been widespread.   

AI is at the core of the effort to control the pandemic and examples are easy to findAI analytics have enabled a faster correlation of data on patients with underlying health conditions beyond diabetes, hypertension and obesity which were identified early on. The power of AI to analyse huge quantities of data quickly and accurately means that doctors now have an expanded list of conditions which increase the risk from Covid-19 and can provide more effective treatments. AI powered algorithms have also been used to predict where the virus will spread and when and, in the US, machine learning is even helping to identify who is most vulnerable and should receive the vaccine as a priority. 

Language AI and healthcare technology; putting the 2 together. 

These are just a few of the pandemic-driven uses of AI in healthcare and the number will undoubtedly continue to grow. But where is language in all this and what part haNLP played? Why is language and in particular multilingualism, important in a medical setting? 

In our blog, Language on the Front Line, we examined how multilingual messaging has been crucial in spreading the correct information about safety and prevention during the pandemic. The crisis has underlined the need to communicate the right message in the right language to reach those most vulnerable, and multilingualism has had a life-saving role. When governments and healthcare providers turn to multilingual AI solutions it is equally important that those solutions are able to manipulate languages expertly enough to avoid misunderstanding and misinformation. There have been many examples of multilingual AI adoption as a result of the coronavirus crisis. 

The German Federal Ministry of Health has partnered with a leading language technology company to ‘facilitate multilingual communication to and from the German language in hospitals and emergency rooms throughout Germany’. This automatic speech recognition and natural language processing technology is being integrated into a telecommunications platform already being used by language mediators and will help to enable urgent conversations between medical staff and non-German speaking patients. A reported insufficient number of skilled interpreters means there is a need to communicate across language barriers in real time and ensure patients receive the necessary and sometimes emergency treatments. 

In the UK, Kettering General Hospital Foundation Trust has begun using live AI translation technology as part of its video consultation platform in an effort to improve accessibility for patients with a lower understanding of English and increase the number of patient-doctor meetings carried out in a virtual setting. The live (machine) translation can be provided in over 100 languages and either in a text or audio format depending on the patient’s preference. The company providing the service, a specialist healthcare communications provider, claims on its website that the hospital trust will save thousands of pounds in traditional payments to telephone and face-to-face interpreters and suggests that the National Health Service would makes savings in the millions if it adopted the system nationwide. 

Multilingual AI chatbots have also been an important tool in combatting the communication difficulties thrown up by the pandemic and they have been used all over the world. One of the most prominent examples is the ‘HealthBuddy’ message bot deployed by the World Health Organisation in Europe and Central Asia and designed to deliver straightforward and accurate messages about Covid19. Its principal objective is to help the wider public sort through the ‘infodemic’ on social media and the internet and is available in over 15 languages from a drop-down menu. 

Can humans be replaced by AI language technology in a medical setting? 

While these uses of multilingual AI have proved effective over the last year, there are also reasons to be cautious. When asked to check symptoms, chatbots can produce inconsistent advice and are heavily reliant on the information they use being updated regularly. AI as we know, is only as good as the data we train it on and providing it with impartial and accurate data is an on-going issue. Putting the ‘multi’ into multilingual can also be a challenge and historically NLP models have been dominated by English and Mandarin Chinese. Whilst this is now changing, the predominance of a handful of languages also leads to less widely spoken languages being under-represented, thus putting those that speak these tongues at a disadvantage when trying to access medical help. Multilingualism in AI means fairer, more inclusive processes. 

But most importantly AI is unable to reproduce the instincts of a human and in the case of healthcare, the added expertise and training of a medical professional. Whilst chatbots can offer better care management and aid patient engagement, it is also possible that they might misinterpret responses, leading to misdiagnosis and failure to give the right advice or treatment. AI is not (yet) able to look into the eyes of a person and get a feel for whether or not they understand the information being given to them or if their answers might be influenced by their anxieties or health issuesTests carried out by an Israeli health tech company have shown that the way a question is framed can have dramatic consequences for patient responses. AI has yet to learn the nuances of rephrasing a question to help the person understand and answer it better. 

Equally the skills of the human interpreter should not be underestimated. These highly qualified professionals offer much more than a simple language transfer. They have specialist training and a high degree of familiarity with medical terminology. They have cultural sensibilities and can read the non-verbal clues which might prompt a reassuring word or an extra question. They help people though traumatic and frightening experiences. Chatbots don’t have this ability.  

Ultimately how and when AI language technology is used in healthcare must be carefully regulated. Despite the immense progress that has been made, AI still has a long way to go before it can truly understand the subtleties of human communication and will never be a replacement for the human touch. 

AI, Language and Healthcare by t'works